This fixes a regression from 6e96ff3e00
where WORDS_BIGENDIAN somehow was confused with WORDS_NEED_ALIGNMENT,
so that setting WORDS_NEED_ALIGNMENT resulted in assuming a big endian
machine. This completely broke the RPi build, which is little endian.
Fixes: 6e96ff3e00 ("Fix ambiguous use of INTEL_SHORT/INTEL_INT")
[Kp: added Fixes tag.]
Querying GL_INDEX_BITS and GL_DOUBLEBUFFER is not allowed in GLES, so
just use a reasonable default values for both. This is only used to
estimate memory consumption by the framebuffer (which is a questionable
approach in itself, since all those are implementation details not exposed
by the GL, and the actual memory used is likely to be higher anyway).
I got a bug report about erros connecting from RPI builds to normal games
on desktop machines. Unfortunately, this patch also changes the savegame
format (which was incompatible to normal desktop builds before, too).
This is discussed in more detail in:
http://www.dxx-rebirth.com/frm/index.php/topic,1986.msg21232.html#msg21232
Integer promotion changed short+short into an int, but INTEL_SHORT only
takes short or unsigned short. Add a cast to force the result back to
short.
Reported-by: derhass <https://github.com/dxx-rebirth/dxx-rebirth/issues/56>
<gcc-4.7 no longer builds due to incomplete decltype handling
>=gcc-4.7 handles -Wredundant-decls correctly
Remove the SConf test and always set -Wredundant-decls.
derhass reports that Raspbian g++-4.8.2 rejects vecmat.cpp for
-Wmaybe-uninitialized on t0, t1. Fix it by adding code to return 0 if
vm_vec_copy_normalize leaves the temporary uninitialized.
Reported-by: derhass <https://github.com/dxx-rebirth/dxx-rebirth/issues/55>
